Complex social systems are composed of interconnected individuals whose interactions result in group behaviors. Sequential decision making in a real-world complex system has many applications, including road traffic management, epidemic prevention, and information dissemination. However, optimizing the decisions in such real-world complex system is difficult because of the high-dimensional and non-linear system dynamics, and the exploding state and action spaces for the decision maker.
Sequential decision making in complex social systems remain a hot research topic and there are many existing works around it. There are mainly two frameworks of that in complex social systems. One is the single-agent framework which view all the components of the system as a whole and optimize the utility function of the entire system. The other one is the multi-agent framework which view each component as a self-interested agent and each agent optimizes its own utility function. The challenges of solving sequential decision making problems in real-world systems are the huge amount of individuals of different species, the high-dimensional state-action spaces, the complex state transitions, the partial observability, the sophisticated spatial-temporal dependences, the lacking a high-fidelity reward function to reflect the real world need, and the expensiveness in interacting with the environment.
We propose to develop reinforcement learning and optimal control algorithms that can effectively and efficiently solve complex social systems decision making problems. For the single-agent framework, we firstly consider a baseline case where the system dynamics and the reward function are known. In this case, we will develop a partially observable discrete event decision process to capture the system dynamics succinctly, and develop a variational inference algorithm with Bethe entropy approximation to tractably solve the optimization problem. We then consider a more realistic scenario with unknown system dynamics and unknown reward functions, which we formulated as an imitation learning problem. To provide better optimization signals and being more robust, we propose to develop a variational kernel learning algorithm to solve this problem. For the multi-agent framework, we consider a real-world scenario where the system dynamics and reward functions are unknown, which we formulated as a multi-agent imitation learning problem. To cope with the exploding dimensionality problem and to maintain scalability, we propose to develop a mean-field kernel algorithm to solve this problem.
|Commitee:||Chen, Changyou, Joseph, Kenny|
|School:||State University of New York at Buffalo|
|Department:||Computer Science and Engineering|
|School Location:||United States -- New York|
|Source:||DAI-B 82/3(E), Dissertation Abstracts International|
|Subjects:||Artificial intelligence, Computer science|
|Keywords:||Deep learning, Imitation learning, Optimal control, Reinforcement learning|
Copyright in each Dissertation and Thesis is retained by the author. All Rights Reserved
The supplemental file or files you are about to download were provided to ProQuest by the author as part of a
dissertation or thesis. The supplemental files are provided "AS IS" without warranty. ProQuest is not responsible for the
content, format or impact on the supplemental file(s) on our system. in some cases, the file type may be unknown or
may be a .exe file. We recommend caution as you open such files.
Copyright of the original materials contained in the supplemental file is retained by the author and your access to the
supplemental files is subject to the ProQuest Terms and Conditions of use.
Depending on the size of the file(s) you are downloading, the system may take some time to download them. Please be